基本信息
浏览量:51
职业迁徙
个人简介
I am a Tenure-track Associate Professor in the Institute of Natural Sciences and the School of Mathematical Sciences at Shanghai Jiao Tong University, working on deep learning theory and computational neuroscience. I obtained my B.S. in Physics (Zhiyuan College) and a Ph.D. degree in Mathematics from Shanghai Jiao Tong University in China under the supervision of Profs. David Cai and Douglas Zhou. Before joining Shanghai Jiao Tong University, I worked as a Postdoc at New York University Abu Dhabi and Courant Institute from 2016 to 2019.
I am interested in understanding deep learning. In both theory and simulations, we found a Frequency Principle (F-Principle) that deep neural networks (DNNs) often capture target functions from low frequency to high frequency in order during the training. This is significantly different from many iterative schemes in numerical analysis, such as Jacobi method. Our theory provides an understanding to why DNNs can have a large capacity to memorize randomly labeled dataset, but still, possess good generalization in real dataset. We also propose an effective model, resembling statistical mechanics, that accurately predicts learning results of extremely over-parameterized two-layer ReLU NNs. The series work of the F-Principle lays a possible new direction for quantitatively understanding deep learning. A summary of the F-Principle can be found here (PDF).
I am also interested in computational neuroscience, ranging from theoretical study and simulation to data analysis. I have collaborated actively with theoretical and experimental neuroscientists.
I am interested in understanding deep learning. In both theory and simulations, we found a Frequency Principle (F-Principle) that deep neural networks (DNNs) often capture target functions from low frequency to high frequency in order during the training. This is significantly different from many iterative schemes in numerical analysis, such as Jacobi method. Our theory provides an understanding to why DNNs can have a large capacity to memorize randomly labeled dataset, but still, possess good generalization in real dataset. We also propose an effective model, resembling statistical mechanics, that accurately predicts learning results of extremely over-parameterized two-layer ReLU NNs. The series work of the F-Principle lays a possible new direction for quantitatively understanding deep learning. A summary of the F-Principle can be found here (PDF).
I am also interested in computational neuroscience, ranging from theoretical study and simulation to data analysis. I have collaborated actively with theoretical and experimental neuroscientists.
研究兴趣
论文共 70 篇作者统计合作学者相似作者
按年份排序按引用量排序主题筛选期刊级别筛选合作者筛选合作机构筛选
时间
引用量
主题
期刊级别
合作者
合作机构
CoRR (2024)
引用0浏览0EI引用
0
0
CoRR (2024)
引用0浏览0EI引用
0
0
CoRR (2024)
引用0浏览0EI引用
0
0
Journal of Computational Physics (2024): 113012
Zhiwei Wang, Yunji Wang,Zhongwang Zhang, Zhangchen Zhou, Hui Jin,Tianyang Hu,Jiacheng Sun,Zhenguo Li,Yaoyu Zhang,Zhi-Qin John Xu
arxiv(2024)
引用0浏览0引用
0
0
COMBUSTION AND FLAME (2024): 113286
加载更多
作者统计
合作学者
合作机构
D-Core
- 合作者
- 学生
- 导师
数据免责声明
页面数据均来自互联网公开来源、合作出版商和通过AI技术自动分析结果,我们不对页面数据的有效性、准确性、正确性、可靠性、完整性和及时性做出任何承诺和保证。若有疑问,可以通过电子邮件方式联系我们:report@aminer.cn